Predicting spontaneous preterm birth in asymptomatic high‐risk women with cervical cerclage

To determine the performance of the predictive markers of spontaneous preterm birth, cervicovaginal quantitative fetal fibronectin (fFN) and cervical length, in asymptomatic high‐risk women with transabdominal, history‐indicated or ultrasound‐indicated cervical cerclage.

ultrasound in the second trimester (16-28 weeks), and quantitative fFN testing from 18 weeks onward. Test performance was analyzed for the prediction of spontaneous preterm birth before 30 weeks (cerclage failure), 34 weeks and 37 weeks, using receiver-operating-characteristics (ROC)-curve analysis.

INTRODUCTION
Cervical cerclage is used to prevent preterm delivery in women with recurrent preterm losses, with or without a short cervix (< 25 mm) on transvaginal ultrasound (TVS) 1 . The population on which cerclage confers clear benefit is still poorly defined, probably owing to the heterogeneity of the group, which includes those with cervical incompetence 2 , structural anomaly 3 , cervical trauma 4 or a history of cervical surgery 5 . Randomized controlled trial data confirm that cerclage has some value compared with conservative management (number needed to treat, 25) 6 ; however, treatment failure does occur and is associated with significant morbidity (cervical trauma, sepsis, hemorrhage) and mortality. In women with recurrent pregnancy loss, the rate of cerclage failure (delivery before 32 weeks' gestation) is 30-40% 7 . Additionally, there is a lack of consensus on the optimal cerclage technique, timing of suture insertion and aftercare.
The impact of cerclage on the predictive validity of quantitative fFN (qfFN) and CL tests is of interest, as stitches may disrupt fFN secretion and/or artificially lengthen the cervix. By surgically closing the cervical canal, fFN concentration in cervicovaginal fluid may be lowered artificially. Alternatively, cerclage may cause a localized inflammatory response, disrupting the maternal choriodecidual tissues and triggering matrix remodeling, with concomitant leakage of fFN into the cervicovaginal fluid 20 . Elongating the cervix with cerclage may confer a reduced risk of sPTB or may represent merely an artifact of intervention, with no reduction in risk.
If predictive utility is retained in women with cerclage, tests could be used to support patient counseling, for example concerning stopping work or admission to hospital, and to inform decision-making with respect to time-sensitive interventions, such as administration of antenatal corticosteroids, in-utero transfer and magnesium sulfate treatment. In this study, we analyzed CL and qfFN data collected prospectively from a large cohort of asymptomatic high-risk women with cerclage. Our aim was to determine the clinical utility of current predictive tests of sPTB when cerclage was placed prior to conception (TAC), in the first trimester (history-indicated) or after cervical shortening (ultrasound-indicated).

METHODS
This was a planned secondary analysis of a prospective cohort of asymptomatic high-risk women presenting to a PTB surveillance clinic in a tertiary referral London hospital over a 6-year  Women were included if they were at high risk for sPTB, received vaginal or abdominal cerclage and did not receive additional prophylactic intervention. High risk was defined as a history of sPTB or preterm prelabor rupture of membranes (PPROM) before 37 weeks' gestation, previous late miscarriage (between 16 + 0 and 23 + 6 weeks) or previous cervical surgery (large loop excision of the transformation zone (LLETZ), cone biopsy or trachelectomy). We excluded women with multiple pregnancy or known fetal congenital abnormality and those who received progesterone or an Arabin pessary, to ensure effects were related solely to cerclage, as well as women who received a cerclage outside of Royal College of Obstetricians and Gynaecologists guidelines.
As part of routine clinical care within the PTB surveillance clinic, all women underwent serial TVS-CL assessment in the second trimester (16-28 weeks' gestation), with qfFN testing from 18 weeks onward, performed by a small number of trained doctors, midwives or sonographers. Routine labor care was provided by clinical staff, and retrospective clinical outcomes were collected by the research team. Surveillance (TVS and qfFN) was performed every 2-4 weeks according to clinical need and as per the clinic surveillance protocol. qfFN samples from women who reported sexual intercourse within 24 h, those with confirmed or suspected rupture of membranes, or those with blood-stained fFN swabs were excluded from the analysis, in accordance with the manufacturer's instructions (Hologic Inc., Danbury, CT, USA).
Women were offered preconception TAC if they had a history of failed vaginal cerclage, defined as spontaneous late miscarriage or sPTB before 28 completed weeks of pregnancy with vaginal cerclage in situ (excluding rescue cerclage procedures, i.e. cerclage inserted with exposed membranes). History-indicated cervical cerclage was offered as per contemporaneous clinical practice, based on the women's obstetric history, and was inserted before 14 weeks. Ultrasound-indicated cerclage was offered to those with a short cervix (< 25 mm) before 24 weeks on ultrasound and with additional risk factors for sPTB.
Maternal demographic data, CL and qfFN measurements and pregnancy outcomes were analyzed. Women were considered to have had a sPTB if they had spontaneous onset of labor or experienced PPROM and delivered Prediction of spontaneous preterm birth in women with cerclage 619 prematurely, regardless of mode of delivery. Women with iatrogenic delivery before the gestational timepoint of interest or incomplete outcome data were excluded from the analysis.

CL measurement
Serial TVS-CL measurements were made in accordance with standardized guidelines by trained operators who fulfilled local governance and assessment requirements. In summary, the woman was asked to empty her bladder, and then the TVS probe was inserted into the anterior fornix of the vagina to obtain a sagittal long-axis view of the echogenic endocervical mucosa along the length of the cervical canal, allowing identification of both the internal and external ora. Taking care not to apply undue pressure on the cervix with the probe to avoid falsely elongating it, the linear distance between the external and internal ora was recorded (in mm) three times over a minimum of 3 min using optimal magnification and zoom settings. Transfundal pressure was exerted for 15 s and subsequent demonstration of a cervical funnel, if present, was noted. The shortest total closed CL of the three measurements was considered for analysis, with short CL defined as one having a length of < 25 mm 16 .

Quantitative fFN measurement
During speculum examination, a polyester swab was inserted into the posterior fornix of the vagina for 10 s to collect a sample of cervicovaginal fluid. The swab was placed into the test buffer solution and analyzed immediately. An aliquot (200 μL) of the sample was analyzed using the quantitative Rapid fFN 10Q analyzer (Hologic Inc.) according to the manufacturer's instructions. All clinicians received appropriate training in the use of the analyzer. Thresholds of 10 ng/mL (lower limit of test) and 200 ng/mL (based on the existing literature 13 ) were predefined. The qfFN assay results were reported in ng/mL and standardized using purified fFN and A128 measurement with an extinction coefficient of 1.28. The reliability of the Rapid 10Q analyzer has been reported previously 21 . For the 10Q Assay, the interassay coefficient of variation is 5.7-7.3% and the intra-assay coefficient of variation is 5.9-7.5%. Experiments performed during product development confirmed a good correlation between enzyme-linked immunosorbent assay and 10Q tests (slope, 0.97; R 2 , 0.82) (pers. comm., Jerome Lapointe, Hologic Inc.).

Statistical analysis
Statistical analysis was performed using Stata version 14.0 (StataCorp., College Station, TX, USA). Data were analyzed for the whole group and according to the type of cerclage. Our primary outcome was sPTB before 30 weeks' gestation and secondary outcomes were sPTB before 34 and before 37 weeks. Descriptive statistics were used to describe the study population, and chi-square test to compare proportions delivered preterm between groups. Receiver-operating-characteristics curves were generated using the first test result after 20 weeks' gestation or following cerclage insertion, as appropriate, and areas under the curve (AUC) were compared between cerclage groups for both CL and qfFN. Predictive statistics were calculated to determine whether CL and qfFN tests after cerclage insertion accurately predicted sPTB. P < 0.05 was considered to indicate statistical significance.

RESULTS
Of 2141 women at high risk of sPTB presenting to the clinic, 258 were identified as having had cerclage. Of these, 77 women were excluded for iatrogenic PTB, incomplete outcome data and/or concurrent progesterone treatment, leaving a study cohort of 181 women (Figure 1).
To ensure that differences in baseline characteristics did not influence our findings, we adjusted for known risk factors for PTB: previous sPTB, late miscarriage, LLETZ/cone biopsy and the number of preterm deliveries at various gestational ages. After adjustment, qfFN ≥ 200 ng/mL (χ 2 = 18.84; P = 0.0008) and CL < 25 mm (χ 2 = 24.27; P < 0.0001) remained significant predictors of sPTB.
Among those with history-indicated cerclage, CL < 25 mm had 100% sensitivity, a negative predictive value of 100% and a positive likelihood ratio of 6.6 for the prediction of sPTB before 30 weeks (Table 3). Overall, the predictive performance of CL for sPTB before 30 weeks was significantly greater (P = 0.0122) in women with history-indicated cerclage (AUC, 0.96 (95% CI, 0.91-1.00)), compared to those with ultrasound-indicated cerclage (AUC, 0.79 (95% CI, 0.66-0.91)), although both showed clinical utility ( Figure 2, Table 2).  Figure 1 Flowchart summarizing inclusion in study of asymptomatic high-risk women with cervical cerclage and occurrence of spontaneous preterm birth (sPTB). *One woman with transabdominal, two with history-indicated and two with ultrasound-indicated cerclage underwent iatrogenic delivery after 34 weeks and were excluded. Data are given as median (interquartile range), n (%) or n/N (%). *One woman with transabdominal, two with history-indicated and two with ultrasound-indicated cerclage underwent iatrogenic delivery after 34 weeks and were excluded. BMI, body mass index; GA, gestational age; PPROM, preterm prelabor rupture of membranes; sPTB, spontaneous preterm birth.

621
The corresponding predictive statistics for qfFN ≥ 200 ng/mL are presented in Table 4. qfFN was shown to be a strong predictor of sPTB before 30 weeks in women with history-indicated cerclage (AUC, 0.91 (95% CI,      Table 2). This study attempted to undertake an evaluation of predictive markers in women with TAC. As no women with TAC delivered before 34 weeks, predictive test statistics were not calculated for this group. However, all qfFN measurements in women with TAC were < 10 ng/mL, giving a specificity of 100%.

DISCUSSION
CL and qfFN are accurate predictors of sPTB in asymptomatic high-risk women with cerclage. CL performs significantly better in women with history-indicated cerclage compared to in those with ultrasound-indicated cerclage. The predictive performance of qfFN was similar between groups, with a trend towards improved prediction in history-indicated cerclage. Clinicians and policymakers should be reassured that tests can be relied upon to discriminate risk and may have utility when counseling women about treatment failure and planning clinical management, with consideration of in-utero transfer and administration of antenatal corticosteroids, to ensure the best outcome for mother and baby. In our cohort, no women with TAC delivered before 34 weeks, therefore we were unable to evaluate predictive markers in this group. However, all women with TAC had normal qfFN measurements, indicating that this test acts as a surrogate for cerclage success. This warrants exploration in future research.

Clinical implications
The predictive value of qfFN and CL has been demonstrated in large prospective cohorts, but these concentrated on asymptomatic singletons with prior PTB, late miscarriage and/or cervical surgery 13,14,22,23 . Women with cerclage are often excluded from prediction studies to ensure that the intervention does not alter the natural course of pregnancy or affect the predictive capability of the test. However, patients who have undergone intervention are at higher risk of sPTB and may benefit most from accurate prediction of the timing of delivery. Removal of the highest-risk patients could also lead to underestimation of the usefulness of predictive tests.
Thus far, evidence for the clinical utility of predictive markers of cerclage failure has been limited and conflicting. In two smaller studies (n = 73 and n = 57) 17,24 , total CL was not related to pregnancy outcome unless other parameters, such as length of cervix above the cerclage or membrane exposure, was noted. We did not have systematic data on these parameters to confirm if prediction could be improved with their inclusion. Hedriana et al. 24 included iatrogenic PTB, PPROM and chorioamnionitis within their PTB outcome; CL is not a powerful predictor of outcome in these situations and may explain why they were not able to show utility. Both studies may have been underpowered and thus unable to show an effect.
Our study supports the findings of Cook et al. 25 and Pils et al. 26 that measurement of CL shortly after cerclage placement accurately predicts sPTB, and that women with cervical shortening after cerclage insertion are at highest risk of sPTB and require close surveillance. Our larger sample of women also confirmed a difference in outcome between those with history-indicated vs ultrasound-indicated cerclage 26 . The mechanism behind this difference in prediction is not clear. Approximately 40% of sPTB cases are preceded by microbial invasion of the intrauterine space. The cervix serves as both a mechanical and chemical barrier to ascending bacteria 27 . Premature cervical remodeling, softening and dilation may result from a proinflammatory cytokine response and prostaglandin release. Ultrasound-indicated cerclage is placed after the cervix has started to shorten, probably in the presence of ongoing inflammation and associated pathophysiological sequelae. The stitch may successfully halt the ongoing pathological process and prevent sPTB but fail to lengthen the cervix beyond 25 mm or reduce fFN concentration, explaining the higher incidence of false positives observed in this study.
To our knowledge, only three studies have been published evaluating the predictive ability of qfFN in asymptomatic women with cerclage 18,19,28 . The authors confirmed that the negative predictive value of qfFN was retained, i.e. the ability of the test to select those patients who will not deliver preterm within a certain timeframe. However, Roman et al. 19 and Duhig et al. 28 also reported an increased rate of false positives within the 4-week window following cerclage. Generally, only patients with ultrasound-indicated cerclage would undergo qfFN surveillance during this window 19 , supporting the hypothesis that cerclage causes a localized inflammatory response, disrupting maternal choriodecidual tissues and triggering matrix remodeling, with leakage of fFN into the cervicovaginal fluid 20 . Braided cerclage has been associated with increased cervical vascularization 29 , which could also explain the increased rate of false-positive tests.

Strengths and limitations
Women and healthcare providers were not blinded to CL and qfFN measurements in this study. As women in the cohort already had cerclage, we believe that further intervention is unlikely to have influenced the timing of delivery, although morbidity and mortality could have been affected. Labor care was provided by clinical staff, and retrospective outcomes were collected by research midwives not working in the clinic, with a low likelihood of bias. However, the findings of this intervention-based analysis in a high-risk cohort with a prespecified intervention should not be extrapolated to other patient groups.
A strength of this study was the heterogeneous make-up of the three cerclage groups, reflecting the patient mix seen in a high-risk PTB surveillance clinic. Predictive markers retained diagnostic accuracy with cerclage in situ in these cohorts. qfFN (P = 0.0008) and CL (P < 0.0001) remained significant predictors of sPTB after adjustment for baseline characteristics and risk factors.
While this large cohort allowed us to draw conclusions about specific types of cerclage, we did not have adequate power to investigate the additive value of qfFN and CL. Future research should seek to understand the synergies between predictive tests, as well as to determine alternative cut-offs and the ideal surveillance window for this population. We have reported previously on the clinical utility of combining predictive markers in a larger cohort 13 ; it is likely that this approach would also have value in women with cerclage.

Conclusions
CL and qfFN were useful markers of sPTB after placement of all types of cerclage, although prediction was most accurate in those with history-indicated stitches. No woman with TAC delivered spontaneously before 34 weeks' gestation. Clinicians should be reassured that these tests can be relied upon to discriminate risk in